EU ASRock Rack B450D4U-V1LQ5 mATX AM4

[RolloZ170]

i guess there is a high possibility that suppressed options are ignored by any CbsSetupDxe.

 

[RolloZ170]

So you mean SuppressIf is a hardcoded Value (in the Firmware File), NOT something that is based on the Register/Variable actual Value ?

if other processors have Cbs menu, why should this be hidden if not because nobody looks at these options(in this BIOS)

 

[luckylinux]

This is interesting! Yes, the original supported CPU list does include G-sku’s for 2000-series, but without graphics actually working. I mean, how could it if the Aspeed is the GPU?
Then maybe, MAYBE, that could mean that 5000-series APUs would also post!? Just without the graphics…

They were only explicit about NO 4000 Series Support in latest BIOS.

EVERYTHING else is up for Interpretation really ...


EDIT 1: actually there are SOME (very FEW) Supermicro Motherboards that I know of (on the Xeon E3 Family) where, even if there is no display output, the APU/iGPU is still WIRED and POWERED. So you could in those rare Cases use it for Plex/Jellyfin/etc Transcoding. But in most Cases if there is no Display Output, then the iGPU is not powered at all by the Chipset IIRC. Just to mention that it's 2 separate Things: a) The Physical Connector and b) The Fact that the iGPU/APU Graphics is powered or not .

 

[RolloZ170]

So you mean SuppressIf is a hardcoded Value (in the Firmware File),

e.g.
EqIdVal QuestionId: 0x127, Value: 0x43B9
Questionsid comes from a test of CPU type

 

[luckylinux]

a single BIOS AGESA can not support two different architectures.

Then why do they claim it supports 3000 AND 5000 Series CPUs ?

Quote:

latest BIOS only supports Matisse and Vermeer CPU(3000 series and 5000 series with no iGPU) So no 4000 series

 

[RolloZ170]

Which is even more weird since the same ID figures in each of the different CBS Files.

the Dxe's are taken dynamicaly based on the CPU type, but use same VarStores.

 

[luckylinux]

Maybe that is where the statement - "in the last BIOS release only 3000 and 5000 series are supported" - holds true

@dbram Did you only have trouble with your 2600X, but not with your three 3000 series CPU?

I'm more puzzled it posts at all than USB not working if that is the Issue.

Technically Speaking I'm NOT on latest BIOS anyways: I'm on 2.09 (as the Motherboard came with), latest is 2.12.

EDIT 1: I mean, I got Network Boot to work, so if it wasn't for the Keyboard not working, I could definitively run an Installer or Shell ...

 

[RolloZ170]

To me it simply matches the actual Value of the Answer to that Question (which we might call "Variable" Value IMHO).

the multiple QuestionId variables are set by BIOS code based on found CPU/RAM/etc.

 

[luckylinux]

It seems there are three ROM-s available for ASPEED:

• VBIOS
• UEFI
• VBIOS+UEFI

Maybe we just need to flash the combined VBIOS+UEFI ROM to ASPEED and then VGA will also work pre-os in legacy BIOS mode.
You can flash in EFI console.

Download link:

https://www.aspeedtech.com/tw/support_driver/

Sure if I could reach and type something in the EFI Shell .

 

[RolloZ170]

THEN show the different Memory Frequency/Timing Settings (e.g. the one at the beginning of this Post). Otherwise do NOT show the Memory Tuning Setting.

sure Questionid's are used for this too.
have tried to run SPR-SP (support up to4800) with 5600 RAM. but limit is set by BIOS code before based on CPU gen (4th=max.4800,5th=max.5600)

 

[RolloZ170]

So I kinda disagree, at least for this Section, that QuestionId matches a particular CPU.

which ? there are dozends of it. this special ID was an expample.

 

[luckylinux]

Regarding no support for 4000 series. Isn’t that series only APUs? Could that be why they say it’s not supported…

I don’t have any APU at my disposal so I couldn’t test unfortunately. But it would be fun if someone could try. For server use it could open up for a more power efficient yet powerful solution. If the iGPU would be accessible then that’s a bonus.

I’ve ordered a Ryzen 7 PRO 3700 on eBay to try out on my board. Will take a while for it to show up unfortunately.

Get some USB3 Header -> Internal Ports Adapter (or a Chassis with one), you will need it most likely.

Not sure if only the AMD 2xxx Series has this Issue but Man, the Rear USB sure is ***** up.

BIOS should be Version 2.09 (Latest is 2.12) if it's like mine.

 

[RolloZ170]

The example I provided in my Previous Post (QuestionId: 0x62).

that works as you explained. butnot all QuestionId are modified just by other options.

 

[RolloZ170]

Is the Not as I think it is ? So Suppress if QuestionId(0x9) != 0xFF ?

correct.
look here: so far i don't know what Questionsflags do.

        SuppressIf
            EqIdVal QuestionId: 0x8, Value: 0xFF
                Not
            End
            Numeric Prompt: "Combo CBS", Help: "No help string", QuestionFlags: 0x10, QuestionId: 0x8, VarStoreId: 0x5000, VarOffset: 0x20, Flags: 0x20, Size: 8, Min: 0x0, Max: 0xFF, Step: 0x0
                Default DefaultId: 0x0 Value: 254
            End
        End

 

[RolloZ170]

I'm really Puzzled that this Motherboard don't support / enable ECC Memory for 5000 Series CPUs where they claim they "support" the CPU ...

poltics. can only be made by the BIOS author.

 

[luckylinux]

What would be lowest possible CPU that will handle full 40G network speed?

Depends ... Mainly on how good you can offload and if you have a NIC (almost mandatory above 25gbps I think) that can do RDMA/ROCE v2.

 

[RolloZ170]

Of course, provided as you said, that there is Hardware hardcoded Supplementary Condition that we CANNOT bypass.

only by reverse engineering the first Dxe which then forwards to the matching special Cbs for the detected CPU type...

 

[RolloZ170]

So you think that there is such a Check in Place after all ?

yes. x86 code ( reverse engineering )

My hypothesis was that I just tried to set the wrong VarOffset for the wrong Dxe File (hence I wanted to try with the other 3 VarOffset Values) ...

i guess nobody listen to that hiddne options later.

But what CPU are you running then ? Or you just gave up on ECC entirely ?

i don't have that board.

 

[luckylinux]

I was planning to use Mellanox CX-3 with Ubuntu 24.04 LTS. Don't have experience with RDMA/ROCE yet. Is it possible to plant this MB with CPU that can handle 40G wihthout RDMA / ROCE?

I gave a try with a Mellanox ConnectX-4 (non-LX) 100gbps QSFP28 in a Supermicro X11SSL-F with an Intel Xeon E3 1230 v5 I think.

It saturated the PCIe Slot Bandwidth at around 32 gbps (I also needed multiple iperf3 Processes in Parallel and add up the Data Transfers.).

Not really an Answer for this specific Question but iperf3 was definitively sweating on that Platform ... NO ROCE or RDMA enabled explicitely and 100% CPU usage on iperf3.

If you look at cpubenchmark and if we can trust it that would probably be comparable to an AMD Ryzen 3 5300U or so

AMD Ryzen 7 5800X vs Ryzen 7 5700X vs Intel Xeon E3-1230 v5 vs Ryzen 5 5600X vs Ryzen 3 5300U [cpubenchmark.net] by PassMark Software

 

[RolloZ170]

I didn't try this Tool before but maybe worth a Shot to see if I missed something: UEFI Editor and GitHub - BoringBoredom/UEFI-Editor: Aptio V UEFI Editor: an alternative to AMIBCP

the usual way to unhide option menu is move them out of the "suppress if... end" clamp.

 

[RolloZ170]

But if you tell me that there is anyway some hardware encoded Part, how would that work ?

same way than BoringBoredom/UEFI-Editor but by hand.
but if no code os looking to that options there will be no effect.
btw: if you enable ECC, have you checked if that setting remained a reboot ? efi shell is too late to enable ECC.

 

[luckylinux]

Looks like I should rather heading for Ryzen 5600X

You asked for the lowest, I gave you my Opinion. You can of course also try to put a 5950x in it ...

Of course my Testing was only iperf3 across 2 Interfaces (on the same NIC Mind you Port 1 to Port 2, hence the PCIe Bandwidth was the Limiting Factor), but it was still full "Load" for the CPU switching between 2 different Linux Network User Namespaces .

EDIT 1: If on top of it you add Firewall Rules, NAT and whatnot, of course you will need a more Powerful CPU

 

[RolloZ170]

Easier would be asking ASRock, but most likely that's also NOT free ...

have some experience with asrock rack. i am done with them.
but bought a SPC741D8UD recently, only reason was the price.
and first BIOS flash destroyed the content of BIOS chip, without valid Intel ME, no BIOS flash after that,
replace BIOS chip with a working one and update BMC FW resolved that.

 

[RolloZ170]

My impression was that setup_var.efi could be impacted when manually changing Settings later on. But I always thought that it survived Reboots.

most options do. but if ECC is not wanted or supported by code its checked at every POST and set according to rules.

 

[RolloZ170]

I'm also NOT sure that when disassemblying with uefitool and ifrextractor it's 100% like it should be (i.e. you can disassemble, modify a bit, then reassemble and it would just work). I fear it's MUCH more complicated than that ...

look at the set of numbers. you can find them with a hex editor in the sct file. you could move a ref. out of the suppress If e.g.

0x1C593         Suppress If: {0A 82}
0x1C595             Variable 0x9 equals 0x2 {12 86 09 00 02 00}
0x1C59B                 Not {17 02}
0x1C59D                 Variable 0x9 equals 0x3 {12 06 09 00 03 00}
0x1C5A3                 Not {17 02}
0x1C5A5                 And {15 02}
0x1C5A7                 Variable 0x9 equals 0x4 {12 06 09 00 04 00}
0x1C5AD                 Not {17 02}
0x1C5AF                 And {15 02}
0x1C5B1             End {29 02}
0x1C5B3             Ref: NTB Common Options, Variable: 0xFFFF {0F 0F 0A 00 0A 00 06 00 00 00 FF FF 00 06 70}
0x1C5C2         End If {29 02}

 

[luckylinux]

It was so nice weather this weekend that i ended up not getting around to it at all
Was 17degrees and only 3m/s on the disc golf course.

I did get as far as to ask for a shipping quote on 2 of these tho.
Pricey at about 10200-10500 area NOK per (with import,shipping etc all included).
View attachment 42732

You want to mount these ASRock Boards and AMD 5700X as "Blades" ?

That's expensive ...

I COMPLETELY blew my Budget for this Month. And of course you can be sure there will be lots of good Deals this Week I will not be able to purchase . I got another 6 x 16GB RAM DIMMs on a Sale though.

 

[Cruzader]

@Cruzader: any Progress on your Side ? This lack of ECC is quite a big Issue IMHO ...

Got the first set of compute nodes moved onto these boards, they are working as expected and have not really looked much more at them beyond that.

 

[RolloZ170]

Uhm but again how would that help with the hardcoded Hardware based Checks ?

not help if options are ignored then. but you see all options.

And how would you rebuild the BIOS after modification ?

UEFITool replace as is

And Flashing must be done via a SOP/SOIP Clip,

yes, without BMC/IPMI the only option

 

[luckylinux]

Yes, thank you for that. I can now make my picture. Since I can buy 5600X cheaper than 5300U, I hope it will work well.

Yeah, I get that. Even if you only had to pay 10 EUR more for it to get 2 Times the Performance it would be worth it .

Be mindful about the PCIe Slot Limitation though. The PCIe 3.0 x8 is only PCIe 3.0 x4 Electrical, so approx (theoretical) 4GB/s or 32gbps.

So you'd need to use the x16 for that if you want full Speed. But then you cannot install NVME and PCIe x4x4x4x4 Bifurcation ...

 

[dbram]

I think 5000 Series CPUs is out of Luck concerning ECC Support. I tried everything I could think of short of Reverse Engineering the X86 Part of the BIOS as @RolloZ170 mentioned (but I don't think I have the Skills to do that TBH).

dmidecode | grep -i "Error correction"

After a fresh Boot you could try grepping dmesg:

dmesg | grep -i edac

If it ONLY reports this, then ECC is not Enabled:

[    0.650269] EDAC MC: Ver: 3.0.0

Thank you

here's mine :
Model name: AMD Ryzen 7 3700X 8-Core Processor

# dmidecode | grep -i "Error correction"
        Error Correction Type: Multi-bit ECC
        Error Correction Type: Multi-bit ECC
        Error Correction Type: Multi-bit ECC
        Error Correction Type: Multi-bit ECC

# dmesg | grep -i edac
[    0.687160] EDAC MC: Ver: 3.0.0
[    4.203283] EDAC MC0: Giving out device to module amd64_edac controller F17h_M70h: DEV 0000:00:18.3 (INTERRUPT)
[    4.203288] EDAC amd64: F17h_M70h detected (node 0).
[    4.203293] EDAC MC: UMC0 chip selects:
[    4.203294] EDAC amd64: MC: 0:     0MB 1:     0MB
[    4.203297] EDAC amd64: MC: 2:  8192MB 3:  8192MB
[    4.203302] EDAC MC: UMC1 chip selects:
[    4.203303] EDAC amd64: MC: 0:     0MB 1:     0MB
[    4.203306] EDAC amd64: MC: 2:  8192MB 3:  8192MB